Gender differences in response to transference interpretations in dynamic psychotherapy

(1)

transference interpretations in dynamic psychotherapy

Randi Ulberg

2010

(2)

© Randi Ulberg, 2010

Series of dissertations submitted to the Faculty of Medicine, University of Oslo No. 1005

ISBN 978-82-8072-518-9

reproduced or transmitted, in any form or by any means, without permission.

Cover: Inger Sandved Anfinsen.

Printed in Norway: AiT e-dit AS.

Produced in co-operation with Unipub.

The thesis is produced by Unipub merely in connection with the

thesis defence. Kindly direct all inquiries regarding the thesis to the copyright holder or the unit which grants the doctorate.

(3)

1 Overview

1.1 Summary

Object: The effect of patient gender in response to psychotherapy has been the object of profound theoretical discussions. In clinical theory, analysts recognize that patient gender contributes to the relationship between the therapist and the patient and the patients thoughts about the therapist. However, the empirical research on whether women or men improve most from psychotherapy has revealed that women and men respond equally well to psychotherapy, i.e. on average, respond similarly across different types of psychotherapy. The empirical explorations of what works for women and what works for men have, however, only sparsely been studied.

Transference interpretation has remained a core ingredient in the psychodynamic tradition, despite limited empirical evidence for its effectiveness. The main aim in the First Experimental Study of Transference-interpretations (FEST) is to explore the long- term effects of transference interpretations in dynamic psychotherapy in the sub groups of patients with high and low Quality of Object Relations (QOR).

The aim in the present thesis is to investigate the effects of transference interpretation in women and men with different levels of relational functioning (high, average and low QOR) during therapy and during follow-up.

The synopsis of this dissertation presents an introduction to dynamic psychotherapy with transference interpretations, research on process and outcome in psychotherapy with a focus on research on transference interpretation, and the effect of patient gender and relational function on outcome. Also the statistical methods used in the studies are described and possible advantages are discussed. A short summary of the individual studies with comments and limitations precedes a further discussion of the main findings, and clinical implications, and some suggestions for future research.

Material and method: Data from the First Experimental Study of Transference- interpretations (FEST) are used. Patients (N=100) were randomized to receive 2 different dynamic psychotherapies during 1 year, with either a moderate level of transference interpretations or no transference interpretations. Follow-ups were at 1 year and 3 years after treatment termination. The outcome measures used were the

Psychodynamic Functioning Scales (PFS), Inventory of Interpersonal Problems –

(6)

Circumplex version (IIP-C), Global Assessment of Functioning (GAF), and Symptom Checklist-90-R (GSI).

In Paper I – IV change is assessed using Linear – Mixed Models (SPSS).

Treatment effect means the effect of transference interpretations. In the statistical analyses in Paper V SABS-works, the SASB-statistical program is also used.

Results: In Paper I the long-term (during the 4 year study period) effects of transference interpretations are explored. Both treatments demonstrate significant improvement during the whole study period. However, patients with a lifelong pattern of poor object relations have a sustained positive treatment effect of transference interpretations (PFS; P < 0.026).

Paper II reports that measured with GSI and GAF (symptomatic, functional change), women respond better than men to dynamic therapy with transference interpretations during therapy. Measured with GAF, patient gender shows moderator effects over and above the moderator effects of QOR (P < 0.03). Women with low QOR show a large positive effect of transference interpretations, but in contrast men with high QOR show a large negative effect.

In Paper III sustained differences in treatment response to transference interpretations between women and men with average relational functioning (QOR) is revealed, measured with PFS (dynamic, interpersonal change). Women and men differ significantly in their response to transference interpretation (P < 0.059). The women with average relational functioning show a long-term positive effect after dynamic psychotherapy with transference interpretations compared to dynamic psychotherapy without transference interpretations (P < 0.037), while the men with average scores on QOR do not.

Paper IV reports sustained differences in treatment response between the two contrasting sub groups of women with poor relational functioning (low QOR) and men with good relational functioning (high QOR) during the whole 4 year study period. Low QOR female patients have a strong positive treatment effect of transference

interpretations (PFS, P < 0.005) while the high QOR male patients show a negative, but not significant treatment effect of transference interpretations.

(7)

In Paper V a highly successful therapy from the sub group of women with poor relational functioning is explored. Qualitative and quantitative data are used. Changes in therapist-patient interaction coincide well with the patient’s sequential improvement.

Conclusions: Both the psychodynamic treatments with and without transference interpretations demonstrate significant improvement during the whole study period.

However, the protocol analysis shows, contrary to the hypothesis, that transference interpretation are more beneficial and have a sustained positive treatment effect for patients with a history of less mature object relations. The post-hoc analyses of gender as moderator show that women responded better than men during therapy with transference interpretations. The difference in treatment effects sustains during follow- up. Female patients, who have difficult relationships to other people is the sub group of patients in FEST showing the best treatment effects from dynamic psychotherapy with transference interpretations.

(8)

1.2 List of papers

Paper I

Høglend P, Bøgwald K-P, Amlo S, Marble A, Ulberg R, Sjaastad MC, Sørbye Ø, Heyerdahl O, Johansson P. Transference interpretations in dynamic psychotherapy: do they really yield sustained effects? Am J Psychiatry 2008;165:763-771.

Paper II

Ulberg R, Johansson P, Marble A, Høglend P. Patient sex as moderator of effects of transference interpretation in a randomized controlled study of dynamic psychotherapy.

Can J Psychiatr 2009;54:78-86. Erratum in: Can J Psychiatry 2009;54:350.

Paper III

Ulberg R, Høglend P, Marble A, Johansson P. Women respond favourably to transference interpretation, men do not: a randomized controlled study of long-term effects of dynamic psychotherapy. (Submitted for publication)

Paper IV

Ulberg R, Marble A, Høglend P. Do gender and level of relational functioning influence the long-term treatment response in dynamic psychotherapy? Nordic J Psychiatr 2009;63:412-419.

Paper V

Ulberg R, Høglend P, Marble A, Sørbye Ø.From submission to autonomy: approaching independent decision making. American J Psychother 2009;63:227-243.

(9)

1.3 Acknowledgements

I am indebted to my supervisor, professor Per Høglend. He has offered me the opportunity to work with the solid data in FEST. With great generosity he introduced me to important objects in psychotherapy research. Framed by a good relational pattern,his advice and support have been crucial for me when working through data;

whatstatistical models to perform and speculate on for whom the results yield sustained effects and how. He has performed profound levels of transference of knowledge and with insight helped me turning unconscious research questions into statistical models;

enhancing autonomy.

In the FEST research group Alice Marble has been of great importance for me during the whole study period. We have been working closely together especially on the SASB-scorings. Paul Johansson has been an excellent fellow researcher and co-author.

We have been sharing many “doctoral ups and downs”. I thank Svein Amlo, Kjell- Petter Bøgwald and Øystein Sørbye for vivid high quality discussions in interpreting the results, and the whole FEST research group including Oscar Heyerdahl and Mary Cosgrove Sjaastad for their contribution in peer supervision and providing treatment data to the study.

I am most grateful for the alliancewith Anne Grete Hersoug and Hanne Sofie Dahl. We share lots of fun while working together. Martin Furan and Martin Nielsen have been most helpful solving numerous practical problems and creating graphics.

Elisabeth Husem has given me important assistance in search of literature and help with the reference lists. I have thrived in the company of the scientists at Institute of

Psychiatry, Vinderen, University of Oslo; always supportive and knowledgeable.

I want to thank Anna von der Lippe and Ken Critchfield for supervision on SASB-scoring, Lars Christian Opdal for discussions on gender differences in clinical practice, and professor of statistics Inge Helland, for supervising the statistical analyses.

While I worked as a child- and adolescent psychiatrist at BUPA, Vestfold, the head of the department Inger Meland Buene, established the framework allowing me to finalise this thesis.

(10)

My parents Kari and Per Ulberg I thank for framing the development of my initial relational patterns and for their perpetual support.

I thank Jan, my love, my life, for sharing with me the everyday dynamics, transference, countertransference,and real relationship far beyond what’s fully interpretable. However, the SASB-scorings are presumably cluster 2, 4, and quite a lot of cluster 3.

Thanks to our children Eigil, Yngve and Endre Michael - the great interactors – making it all worthwhile.

(11)

2 Introduction

2.1 Dynamic psychotherapy – some basic principles

Dynamic psychotherapy can be defined as a treatment that focuses on thoughtful timed interpretations of transference and resistance and a sensitive appreciation of how the therapist contributes to the interaction with the patient (Gabbard, 2004). After Sigmund Freud described his theory about drive, different theoretical models have become useful and fundamental in dynamic psychotherapy; e.g. object-relation theory, self psychology, and attachment theory. Dynamic therapeutic work is based on some key assumptions:

a) Much of a person’s mental life is unconscious.

b) Early childhood experiences in combination with genetic factors both contributes to the development of the person and her or his patterns of object relations (Klein, 1952).

c) The patient’s transference of emotional patterns (representations in the inner world) to the therapist (outer world) is a primary source of understanding.

d) The therapist’s countertransference feelings are a process that might facilitate the understanding of the patient’s inner life (Heiman, 1950; Sandler, 1976).

e) The therapist’s focus on the patient’s resistance to the therapy process is helping, in the working through (Freud, 1914; Røssberg et al., 2003) of emotions and personal patterns.

f) Transference interpretations helps to alleviate psychiatric symptoms.

g) The goal of psychodynamic psychotherapy is sustained structural or psychodynamic change, which means improvement in areas such as tolerance of affects, interpersonal functioning and insight (Gabbard, 2004;

Ulberg, 2008).

The concepts dynamic psychotherapy, psychodynamic psychotherapy and psychoanalytically oriented psychotherapy are often used synonymously. Dynamic psychotherapy may be of short or long duration. What is long-term and what is short- term or brief psychodynamic psychotherapy can be defined differentially. The tradition in Europe seems to have been that psychotherapy less than 1 year has been defined as

(12)

short-term. However, Gabbard (2004) writes that ”….. the definition of long-term is a duration greater than 24 sessions or 6 months.”

The key assumptions in dynamic psychotherapy mentioned above, are applicable independent of treatment length. Some technical innovations in psychodynamic

psychotherapy of short to moderate length are:

a) A focus should be negotiated.

b) Greater therapist activity.

c) The therapist should encourage problem-solving strategies during and after therapy.

d) Patients should be instructed about the principles and procedures of dynamic psychotherapy (Høglend, 1996).

2.2 Transference and transference interpretations

Transference interpretations are commonly understood to refer to the therapist making an explicit reference to the patient’s reaction to her or him. The patient’s reaction to the therapist is to some extent determined by the patient’s previous relationships (Piper et al., 1991). Emphasis on transference interpretation is a hallmark of dynamic psychotherapy. This technique distinguishes this treatment modality from other forms of psychotherapy.

Sigmund Freud first described clinical transference enactments and the

interpretations of them more than one hundred years ago. The first detailed clinical case describing transference in treatment was the case study called Dora (Freud, 1905).

Freud discovered that feelings in psychotherapy were transferred from early childhood relationships and experiences (übertragung). According to Rosenbluth (1961) Freud described that children relate to the mother’s breast (i.e. the child relates to an object).

This thinking seems to precede the understanding of the relational structure of the inner world. Later theorists have further developed the relational understanding (e.g. object relation theory [Klein, 1952]). The internalization of relational experiences occurs through three mechanisms: incorporation, introjection, and identification (Piper and Duncan, 1999). The child uses different mechanisms at different development levels:

(13)

a) Incorporation occurs during the early stages of development. I.e. during the symbiotic/preverbal age when there is confusion about what is oneself and what is the other.

b) Introjective processes require greater differentiation between self and object.

c) Identification is the process where aspects of the other are assimilated and transformed, wholly or partially (Laplanche and Pontalis, 1973/2006).

When integrated with parts of the self, the internalized aspects contribute directly to the establishment of a core sense of identity and inner object relations. Thus, the inner patterns are at least partly based on experiences of the early important persons and relations in the patient’s life. The person transfers her or his object relational patterns, to other persons outside therapy, and in therapy, to the therapist. One should remember, however, that the real relationship between the patient and the therapist, and the therapist’s behavior also contributes to the patient’s reactions to the therapist.

Interpretation of the patient’s maladaptive relational patterns (relational interpretations) is the primary technique used to increase self-understanding and relieve psychiatric symptoms. However, there are largely varied definitions of what exactly interpretations are and how they may be classified. In the psychodynamic tradition there seems to be a general agreement that interpretations include interventions that aim to establish connections (by use or analogy) between different internal dynamic components (e.g. wishes, needs, motives, affects, defense, and anxiety) and past or present objects (i.e. persons and therapists). Transference interpretation is a subtype of relational interpretations that emphasize to help the patient understand her or his relational reactions and behavior within the therapeutic relationship (Gibbons et al., 2004). Those interventions explicitly address the dynamics of the patient’s behavior toward and experience of the therapist in the here and now (Stone, 1981; Gabbard, 2004; Høglend, 1990). A focus on the conflicts and themes that arise in the therapeutic relationship will have immediate affective resonance and illuminate the “true” nature of problems in the patient’s relationships outside therapy. Transference interpretations are thought to set in motion a chain of events that bring about insight and change that will protect against future stressful events, and also enable the patient to make better plans for the future (Gill, 1979; Høglend, 2004; Paper V). Exploration of the patient’s

(14)

transferential reactions to the therapist, are unique opportunities for insight and psychic change.

Various definitions of transference interpretations have been proposed. It is imperative when using this term, both in research and in clinical case conferences, to identify which definition of transference interpretation is employed. The most clear cut difference across authors is that for some authors any discussion of the therapeutic relationship qualifies as transference interpretations, while others classify interventions as transference interpretations only if there is a historical/genetic link made as well. The genetic interpretations seem to be infrequently used (Piper, 1991; Høglend, 1993).

A narrow definition linking genetic interpretations of unconscious material to transferred aspects of the patient's perception of the patient-therapist interaction, can probably not be used by researchers. The distinction between unconscious and conscious and between what is transferrential and what is a realistic perception of the therapist is very difficult to draw.

In FEST (Høglend et al., 2006) the transference work was categorized in different levels. The levels represent degrees of comprehension from superficial and preparatory (Level 1-3) to profound (Level 4-5) analysis of the emotions and behavioral patterns:

Level 1: the therapist addressed transactions in the patient-therapist relationship.

Level 2: the therapist encouraged exploration of thoughts and feelings about the therapy and the therapist, including repercussions to the transference by high therapist activity.

Level 3: the therapist encouraged the patient to discuss how the patient believed the therapist might feel or think about the patient.

Level 4: the therapist included herself or himself explicitly in interpretive linking of dynamic elements (conflicts), direct manifestations of transference, and allusions to the transference.

Level 5: the therapist interpreted repetitive interpersonal patterns and linked these patterns to transactions between the patient and the therapist.

For examples of transference interpretations and work in the transference, see the clinical vignettes in Paper I and Paper IV and the case study in Paper V.

(15)

2.3 Research on process and outcome in psychotherapy

The field of psychotherapy research has made great advances in the last decades.

It supplements the theory-based activities of therapists and provides a foundation for practice (Lambert et al., 2004). The aim of psychotherapy research is to advance our knowledge about the process as well as the course and the outcome of psychotherapy.

Process refers to what happens in psychotherapy sessions (e.g. therapist interpretations).

Outcome refers to short- or long-term changes that occur as results of therapy (e.g.

symptom relief, dynamic change) (Hill and Lambert, 2004). The ultimate goal is to identify the best treatment options possible for patients with a given problem, disorder, or set of problems and study through which mechanisms the therapy works; what works for whom and how (Roth and Fonagy, 2004)

2.3.1 Common versus specific factors

Reviews of the literature have provided support for the effectiveness of psychotherapy approaches to the treatment of mental disorders, (Leichsenring et al., 2004; Joyce et al., 2006). When comparing the efficacy between different

psychotherapy modes, no psychotherapy type seems to be more helpful than the others.

Possible reasons for this have been discussed. It might be that the effect of therapy depends on common factors similar across treatment types (Lambert, 2004). Some elements are regarded as being common to effective therapies (e.g. patient expectations, identification with the therapist, therapist empathy, a good working relationship between the patient and the therapist, and corrective emotional experiences). However, it might be that studies have lacked precision to detect differences. Different treatments might achieve on average, similar outcomes through different processes for different patients. Some researchers emphasize the need for mode-specific outcome scales (Høglend et al, 2000) measuring the expected specific change for different therapy modes. The specific factors perspective advocates the need for knowledge about specific techniques that cause patient improvement, and for whom and how it works.

2.3.2 Study designs

A variety of methodology approaches (e.g. single-case studies,

naturalistic/observational studies, randomized controlled studies, and dismantling

(16)

designs) might be used when investigating the process and outcome in psychological treatments. The methods can be ranged hierarchically depending on their capacity to detect causal association between therapy process and outcome.

Case studies

Single case studies represent a method with low capacity in testing causal relationships between process and outcome. Development in psychodynamic psychotherapy has for a long time, been based on the use of case reports and case studies. Traditional case studies have been criticized for relying too heavily on narratives (Lorentzen and Høglend, 2002; Paper V). The data from one patient are not compared with results from other patients and significance tests can not be computed.

Thus a single case study can not make basis for causal conclusions. To improve the case study design, Hillard (1993) described three basic types of single-case research: A) Single-case experiments that involve quantitative data, manipulation of treatment variables, and hypothesis testing. B) Single-case quantitative analyses that might be used for generation of hypotheses. C) Single-case studies using qualitative data. By mixing quantitative and qualitative designs, one might to some extent, compensate for the subjectivity in traditional case studies.

Naturalistic/observational studies

Naturalistic studies (often imprecisely referred to as effectiveness studies) are one group studies of the natural course of therapies. As no randomization or

experimental control is performed, internal validity might be compromised. Naturalistic studies focuses on external validity and the generalizability to “real-world” conditions.

This design emphasizes implementation of the therapeutic procedure to general practice (Johansson, 2008; Kendall et al., 2004). Change might be studied, but only correlational analyses can be performed. Significant correlations, however, are not proof of causal effects.

Randomized controlled trials

In a randomized controlled study the included patients are randomized to a study treatment condition or to a different treatment condition/placebo. The response in a

(17)

treatment group is compared to the response in a control condition. The efficacy of a treatment might thus be determined. Patients and evaluators should be blind to treatment group. Efficacy studies emphasize the internal validity:

a) Controlling the types of patients included in the study.

b) Using manuals to standardize the treatment.

c) Training the therapists prior to the study.

d) Supervising the therapist during treatment.

e) Monitoring the technique adherence and the “dose” of the therapy.

f) Random assignment to treatments.

g) Prior selection of the primary outcome measure(s).

To determine whether a treatment is efficacious or not, simply involves a demonstration of the treatment being significantly superior to the effects of a placebo condition, some minimal treatment or waitlist. Trying to use placebo conditions in psychotherapy research is associated with numerous methodological and ethical problems (Borcovec and Sibrava, 2005). Placebo is thought to contain no ingredients that target the problem being treated. The placebo effect is psychological. In

psychotherapy all changes are due to psychological factors. Therefore, it is hardly possible to create a psychological control procedure that is inactive. Minimal treatment is usually less credible to the patients. Waitlists may cause demoralization.

Dismantling design

Borcovec and Sibrava (2005) discuss that even the RCT method applied in psychotherapy research might ignore to study what specific therapist behavior causes the change in the patient during and after therapy. They maintain that psychotherapy studies are mostly comparing therapy modes that differ in a number of ways. Therefore RCTs often do not generate significant basic knowledge.

Borcovec and Sibrava’s advice is to focus on identification of putative specific techniques that cause patient improvement. Studies should investigate treatment conditions that differ in only one component. Dismantling or component control designs contrast the complete therapy (e.g. dynamic psychotherapy with transference

interpretations) with the same therapy with experimentally manipulated differences solely in one single dimension (e.g. dynamic psychotherapy without transference

(18)

interpretations). If also the quality of the patient-therapist relationship in both conditions is equal, therapist effects are minimized, and the quality with which interventions are given is equal in both groups, then the study offers the best possibility to detect causal relationships between the specific technique and outcome.

2.3.3 Validity

To know definitely and beyond doubt whether a psychotherapeutic method works, the validity of the measures used also needs to be considered. Validity is the ability of a measure to estimate or describe the phenomenon or construct it purports to do. Shadish and colleagues (2002; Lund, 1996) developed a validation system based on three suppositions: cause comes prior to effect, there is a relation between cause and effect, and other explanations for the effect are excluded. Shadish and colleagues (2002) described four types of validity:

1. Statistical validity refers to the question of whether there is a statistically significant relationship between the independent and dependent variable, i.e.

treatment and outcome (e.g. transference interpretations and therapy outcome).

Significance level balances between Type I errors and Type II errors. Type I error is the probability of rejecting a true null hypotheses (false positive, finding differences (effects) that are not true in the population). Type II errors is the error of failing to reject false null hypotheses (false negative) (Cohen and Cohen, 1983).

E.g.: In FEST the significance level of 0.10 was decided à priori for the moderator analyses and the sub group analyses in order to balance the risk of Type I and Type II errors.

2. Internal validity refers to the causal inference between treatment and outcome.

Threats to internal validity especially in naturalistic studies with no control group, might be:

a) Maturation or spontaneous remission. Naturally occurring changes over time could be confused with a treatment effect. E.g. the patient becomes wiser, older and stronger (Lorentzen, 2002).

b) History. Other factors than treatment have led to the observed change.

(19)

c) Testing. Repeatedly use of a measurement instrument influence the patients’ response.

d) Instrumentation. When the raters use an instrument repeatedly, the raters get more experienced and rate in a new way.

e) Selection of the patients to the study.

f) Regression to the mean. A statistical phenomenon due to

measurement error which may create the illusion of improvement in a treatment study. A precondition of the phenomenon is that the patients are selected on one pre-treatment variable.

In a randomized controlled study threats to internal validity could be:

a) Drop-out. The drop-out could differ between the treatment group and the control group. In FEST the drop-outs were included in the intention-to-treat outcome analyses. There was no drop-out at 3 year follow-up evaluation in FEST.

b) Atypical behavior in the control group. In long-term follow-up studies differences in positive and negative life events and additional treatment might occur. In FEST the patients were advised not to receive additional therapy in the first year after therapy. Within the sub-samples of low QOR patients 15% in the transference group but 55% in the comparison group consulted mental health professionals during the 2 last years of the follow-up (Johansson et al., In Press).

3. External validity concerns the generalizability across different patient samples and settings. Naturalistic studies may emphasize external validity, while in a randomized controlled trial with high internal validity; the external validity might be threatened. The patients may have been selected from a narrow specter of diagnoses, and a strictly manualized treatment will no longer represent treatments as it is carried out in clinical practice. Thus, FEST differs from ordinary practice. However, trying to compensate, the patients included in the study sought psychotherapy for a broad specter of psychiatric symptoms, and were referred from general practice, outpatient clinics and private practice.

4. Construct validity refers to the degree to which the operationalized independent and dependent variables represent the construct of interest. If a beneficial

(20)

outcome is defined as symptom reduction or improved interpersonal functioning, the instrument used to assess outcome should measure just that.

Threats to construct validity might occur:

a) True changes in the construct of interest are not detected.

E.g.: Self-report scales may fail in differentiating people who are healthy from those who are defensive deniers of distress (Shedler et al., 1993) and may be less sensitive in detecting change than clinician- rated scales.To counteract the possible effect when using self-rating scales (IIP-C, GSI) in FEST, two clinician rated-scales (PFS, GAF) were applied. These scales were rated with high precision using three expert raters on each occasion.

b) Other irrelevant constructs are being measured.

c) The patients guess the hypothesis and report improvement accordingly (demand characteristics).

d) Experimenter expectations (researcher allegiance). The experimenter can influence participant responses by conveying expectations about desirable responses, and those expectations are part of the treatment construct tested (i.g. treatment format may favour one treatment, outcome measures may be more sensitive to one treatment, etc.).

2.3.4 Reliability

Conclusions of high validity are dependent on reliable measures. Reliability estimates describe the consistency of a measure. The reliability can be estimated in different ways:

a) Test-retest reliability of an instrument: the instrument is applied repeatedly at different time points. Stable traits (e.g. intelligence) will show high test- retest reliability while state-like conditions (e.g. depression) will show lower test-retest reliability.

b) Interrater reliability can be used to measure the degree of agreement between two or more raters. Estimation of interrater reliability can be performed in different ways. For categories (e.g. SASB-clusters), Kappa might be computed. Kappa is the number of agreements divided by total

(21)

number of observations, corrected for differences in agreement. If including degree of disagreement (e.g. whether the ratings were one or more clusters apart in SASB, Appendix 8.1), weighted Kappa is obtained. For continuous data, Person correlations has often been used to establish the extent of consensus between two observers ( Lorentzedn, 2002).

For continuous scales (e.g. PFS) Intraclass Correlations (ICC) are more accurate. They are computed by comparing variance components across patients and raters:

Reliability (ICC) = Patient variance___________________

Patient variance + Rater variance + Residual variance

c) Internal consistency reliability, usually called Cronbach’s alpha, estimates the consistency across items on the same test, and is used with measures that have several items. Split-half correlations are performed where the sample is split in two in all possible ways, and the sub-samples are correlated. The coefficient is the average of all these coefficients.

2.4 Measurement of change

Measurement of change comprises a variety of methods. With post-treatment measures (e.g. post-treatment benefit rating) the patient and/or the therapist evaluate the degree of improvement based on subjective and/or clinical impression. Different statistical models can be used. Independent sample t-test and One-Way ANOVA test differences between treatment groups at post-treatment. Ordinary Least Squares- regression (e.g. ANCOVA, Multiple regression) are used to estimate pre-to post- treatment change. Repeated measures General Linear Models (GLM) and Growth Curve estimations (e.g. Multi Level Models) are used to estimate change over time. Clinically significant change should also be considered to decide whether a therapy or treatment is not only statistically effective, but also clinically effective. Effect Size estimates are standardized differences, within or between groups, in order to have benchmarks for the

“magnitude” of effects (Cohen, 1988).

(22)

2.4.1 Clinically significant change

Some authors argue that psychotherapy researchers should use Jacobson and Truax’s (1991) method as a standard method to estimate clinically significant change (Lambert and Ogles, 2009). Clinically significant change has to do with; a) the patients’

change need to be larger than measurement error (i.e. treated clients have made statistically reliable improvements from pre- to post-treatment) b) return to normal functioning. A cutoff point between a patient (dysfunctional) and a non patient (functional) population is estimated. Patients who meet both criteria are classified as changed to a clinically significant degree (Jacobson and Truax, 1991). For significant change in outcome measures used in FEST, see Appendix 8.1 on outcome measures.

2.4.2 Statistical Models

Statistical models are mathematical representations of population behavior. They describe features of the hypothesized process of interest among individuals in the targeted population. When using a statistical model to analyze a particular set of data (sample), one aims to find the “true” estimates in the background population. Ideally the sample should be randomly drawn from the population of interest.

A variety of modern statistical models can be helpful to study the complex relations between the patient, the therapist, the process in therapy, external events in the life of the patient, in-session progress, post-session progress, and therapy outcome at the end of treatment as well as during the follow-up period. Recent developments in statistical models have improved psychotherapy research. For a long time research was limited to demonstrate the average difference between two groups using only pre-post change (Lutz and Hill 2009; Pallant, 2001; Norusis, 2004).

2.4.3 General Linear Models

The t-test, analysis of variance and covariance, and regression analysis are all special cases of General Linear Model (Norusis, 2004). In GLM a normally distributed dependent variable is predicted from a linear combination of independent variables.

When using GLM it is assumed that all observations are independent and have a constant variance. However, this assumption of independence might often be violated.

When multiple measurements over time on the same subject are performed (e.g. in

(23)

FEST repeated measurements on different time points for each patient), observations from the same subject are not independent.

Variance is a measure of the amount of variation within the observed values of a variable. Analysis of variance is used to test the hypothesis that several means are equal (H0). Regression or analysis-of variance-techniques are commonly used to test

hypotheses about the relationship between the dependent and one or more independent variables. Analysis of variance techniques, including One-Way ANOVA, repeated measures ANOVA, and analysis of covariance (ANCOVA) have been the mostly used methods. ANCOVA is commonly used to assess post-treatment differences while controlling for pre-treatment levels. ANCOVA fits a single regression line to the relationship between pre- and post-treatment scores for all participants. This regression line is used to test H0-hypothesis; i.e. whether the average regression lines run parallell.

There are different problems for psychotherapy researchers with ANOVA and ANCOVA (Singer and Willett, 2003; Norusis, 2004; Tasca and Gallop, 2009). Both methods emphasize group means and variances, and each individual must have complete data at all time points.

Some difficulties when using ANOVA and ANCOVA:

a) Non linear true change across time may yield inaccurate estimates of the effect of initial levels.

b) Because of multiple tests of the data, accumulating Type I error requires correction.

c) The methods comprise a relatively simple covariance structure and assume constant variability on individuals across time. This assumption is almost always violated.

2.4.4 Linear-Mixed Models

Multilevel models are more sophisticated quantitative methods than General Linear Models and allow longitudinal analysis with the possibility to integrate several levels of changes (e.g., individual trajectories, treatment groups). The multilevel models provide the flexibility of modeling not only the means of the data but the variances and a larger number of covariance structures as well (SPSS 16, 2007; Singer and Willett 2003; Norusis, 2004). Examples of software are Linear-Mixed Models (LMM),

(24)

Hierarchical Models, and Mixed Linear Models. Multilevel models expand the GLM so that the data are permitted to exhibit correlated and non constant variability. However, multilevel models also assume normality of data distribution. Despite multilevel models allow for missing data and nonparallel waves of data across individuals, the design must make sense for the study objectives. Ex.: If in a psychotherapy study the control group received 10 and the treatment group received 40 therapy sessions over one year it would not be appropriate to compare the two groups.

Satisfying models for Effect Size estimates are still not developed. However, Confidence Interval (CI) provides a measure for the precision of the effects found when using multilevel models.

Linear-Mixed Models which are the multilevel models used in FEST, allow researchers to model individual change and variance as well as group change models.

LMM allow the researchers to assess the trajectory (shape) of within-person growth, or change over time, and also between-person differences in growth or change over time, and explain or predict between-person differences in growth or change over time. LMM provide new opportunities for handling missing data in longitudinal designs as well as in nested designs. One can model change even if some individuals have incomplete data without resorting to listwise deletion or imputation of data, as well as non constant times at which data values are obtained. The assumption is that data are missing at random. The LMM allow assessment of whether important estimates (i.e. change in outcome over time per condition) are dependent on missing data patterns (i.e.

informative). Separate treatment effects are estimated for specified missing data patterns, and then an overall treatment effect is calculated as a weighted average of the treatment difference over the patterns (Norusis, 2004; Singer and Willett, 2003; Gallop and Tasca, 2009). With LMM also nonlinear change in individuals can be modeled.

LMM provide an opportunity to model dynamic fluctuation in individual data across time.

Some terms frequently used in Linear-Mixed Models

Subjectis variables that define the cluster of observations. For example in the LMM-analyses in FEST, the patient is defined as the subject.

(25)

Independent variables are defined as factors (categorical variables dividing the observations into groups) or covariates (ordinal measures) Factors/covariates can be further classified as fixed or random. A study can have both fixed and random factors.

The statistical model for the data is then called a mixed model.

Fixed effects have a single constant value for all units of the sample or

individuals. For fixed effects, the hypothesis about means of the dependent variable for the various levels of the effect is tested. In FEST treatment, QOR and gender were treated as fixed effects only. Intercept and time were treated as fixed effects and as random factors. The fixed effects were the test of the average intercepts and slopes between the two treatments, random factors allow each participant to vary around those averages. Thus the variance of the effects and fixed factor parameters are estimated and tested. For example in FEST, initial scores on PFS (intercept) and PFS growth rate across the whole treatment period (slope) of the dependent variable are assumed to vary randomly between individuals (subjects) and the randomly distributed intercepts and slopes were fitted for each patient. So individuals’ intercepts and slopes each have variance associated with them. To determine which random effects should be used in the analysis, it is necessary to describe, the variance/covariance structure of the data.

For example, since the variance between therapists in FEST was almost zero, therapist was not included as a random effect.

2-log likelihood is one of the measures that can be used to compare the goodness of fit of different statistical models. Models with smaller goodness of fit are better. For example, in Paper I, III, and IV log transformation of time was chosen because it fitted the data better than did a linear time slope.

Multilevel models for change address within-person and between-person questions about change simultaneously. Level-1 of the model is associated with intraindividual change (Singer and Willett, 2003). Level-2 of the model is associated with interindividual change. One might specify the multilevel model for change by postulating a pair of subsidiary models (Figure 1): 1. A level-1 submodel that describes how each person changes over time; in the level-1 submodel within-person changes are calculated. 2. A level-2 submodel that describes how these changes differs across subjects; in the level-2 submodel between-person changes are calculated. There might be more than two levels. E.g. in FEST there is a level-1 within-patient change, could

(26)

have been a level-2 between-therapist change and a level-3 between-therapy change.

However, the variance between therapists where almost zero so the model used in the analyses where the level-1 within patient and the level-2 submodel between therapies (Figure 1).

Level-1 submodel

In the level-1 submodel for individual change, the within-person or individual change, is tested with two unconditional models. These unconditional models

decompose the total variation in two different ways. First across people without regard to time (the unconditional means model) and second, across both people and time (the unconditional growth model). The results of these calculations, establish whether there is a systematic variation in the outcome that is worth exploring, and how much variation is found within or between people.

The unconditional means model: In Paper I the main hypothesis was that the transference group would have a more favorable course over the entire study period (4 years) than would the comparison group, as reflected in scores on for instance PFS.

When the research question is articulated it might be tempting to begin by fitting models that include the substantive predictors (e.g. treatment mode). However, one should first fit the two simpler (unconditional) models. The first step is the

unconditional means model. Instead of describing change in the outcome over time, it describes outcome variation acrossall time points and subjects:

Yij = oi + ij

oi = oo + oi

oi is the true mean for Yi (the intercept as a fixed effect), i.e. oi is the person specific mean.

Y_ijdeviates from individual_i’s true mean oi by ij. The level-1 residual is thus a

“within-person” deviation that assesses the “distance” between Yij and oi. oo is the true mean (the grand mean) across everyone in the population. For person i the true mean (oi) deviates from the population average true mean (oo) by oi. ij and oi are

(27)

the residuals (random effects). Since time is not a variable in this unconditional level analysis all time points are included in this model.

-1

An example from FEST of the unconditional means model with PFS as the outcome variable:

The estimated grand mean (oo) is 68.9. The residual variance (oi) is 26.8. The variance of person specific true mean (oi) is 13.7 (P < 0.00, CI 90%; 9.5 – 19.8). Thus half the variance is person specific variance.

The unconditional growth model: The unconditional growth model represents the expected change (slope) each member of the population will experience during the time period under the study.

Yij = oi + 1iTIMEij + ij

oi = oo + oi

1i = 1o + 1i

oi represents individual i’s true initial status, the value of the outcome when TIMEij = 0 (intercept).

1i represents individual i’s true rate of change during the period under the study (coefficient for slope)

ij represents that portion of individual i’s outcome that is unpredicted on occasion j. Ex.: In Paper I, the level-1 growth model represent the individual change on PFS that is hypothesized would occur during each patient’s therapy and during the follow up period:

The PFS estimated grand mean (oo) is 64.0 (CI 90 %; 63.2 - 64.8).

How much of the total variance is explained by time variance might be calculated:

1i Time Variance 1.55 _______ = _____________________________________ ______ = __________________ = 0.06

ij + oi + 1i Residual Variance+ Intercept Variance+ Time Variance 11.0 + 15.45 + 1.55

(28)

6% of the total variance is explained by time. The variance between slopes is significantly different from zero, which means that slope can be used as outcome variable. This is level-1 variation. The residual variance means that there is considerable variance left to be predicted at level-2, for example by the fixed factors treatment (treatment group), or QOR, or gender.

Level-2 submodel

The individual growth parameters of the level-1 submodel are the outcomes (dependent variables) of the level-2 submodel. The level-2 submodel is a model for interindividual change. Treatment is a level-2 fixed effect.

oi = oo + o1TREATMENTi + oi

1i = 1o + 11TREATMENTi + 1i

-oi is the person specific mean of intercept at time 0 (level-2 specification for the level- 1 intercept).

- _1iis the true person specific rate of change (level-2 specification for the level-1 slope).

-_oorepresent the initial status (grand mean).

-1o represent the population average rate of change.

-o1 represent the effect of treatment on level of outcome.

-11 represent the difference in average slope from treatment 0 to treatment 1.

-oi is a residual that assess the distance between the person specific mean (oi) and the grand mean (oo).

-_1iis a residual that assess the distance between the true individual rate of change (1i) and the population average rate of change (_1o).

(29)

Composite multilevel model

SPSS software program collapse the level-1 and level-2 submodels together algebraically into a single composite multilevel model.

Y_ij= oi + 1iTIME_ij + _ij= (

Y

]

oo + o1TREATMENTi + oi) + (1o + 11TREATMENTi + 1i)TIMEij + ij

The first parenthesis contains the level-2 specification for the leve-1 intercept, oi ; the second parenthesis contains the level-2 specification for the level-1 slope, 1i. Multiplying out and rearranging terms then yields the composit multilevel model for change:

ij = [oo + 10TIMEij + o1TREATMENTi + 11 (TIMEij X TREATMENTi)]

+ [oi + 1iTIMEij + ij]

In FEST (Paper I, Table 3) the term (o1TREATMENTi) is not used in the statistical analyzes. If o1TREATMENTi is removed from the model with the interaction still in, we force both treatment to have a common intercept. (Fitzmaurice, 2004). In

randomized controlled trials this is advisable because we can assume that both treatments starts at the same intercept value – which they do in FEST. This model is more powerful and should be routinely used:

Yij = [oo + 1oTIMEij + 11 (TIMEij X TREATMENTi)] + [oi + 1iTIMEij + ij

Ex.: PFS = intercept + time + time x treatment + residuals

To explore the effects of moderators (i.e. QOR, gender) in FEST, this model was used:

Yij = [oo + 1oTIMEij + 11 (TIMEij X TREATMENTi) + 02 MODERATORi + ₁₂ (TIME_ij X TREATMENT_i X MODERATOR_i)] + [_oi+ _1iTIME_ij+ ij].

Ex.: PFS = intercept + time + time x treatment + QOR + time x treatment x QOR + residuals

(30)

Nesting:Comparing PFS from different patients in level-1 means comparing independent data. However, comparing PFS from the same patient at different time points, means comparing dependent data. I.e. time is nested in one single patient and again the patients (within person change in level-1) are nested in the two treatment modes in level-2 (Figure 1). Each patient’s level of independence differs throughout the study period.

Degrees of freedom in LMM are the estimations of the subject’s level of independence on each time point. E.g. In FEST 99 patients are measured with PFS at 4 time points; N400. When analyzing measures from each patient at different time points (within person change), the estimates are nested within each patient and then the data are dependent. When comparing and analyzing data from different patients at one time point, the data are independent. When comparing data from different patients over time (between person change), the data will be dependent within person, but independent between person. In FEST the data are also nested within therapy mode. Thus the degrees of freedom must be estimated and explains the relatively unusual df – calculations; e.g. 130.537, 116.299 etc. (Paper I, Table 3)

In LMM centering might provide stability in the estimation of the chosen statistical model. Centering also improve the interpretations of predictors and moderators. The primary rationale for centering is that it simplifies interpretation. To center the moderators before analysis, provides possibilities to do direct interpretations of parameters. Moderators can be centered at overall means or predefined cut off scores (i.e QOR 5 in FEST, Paper III, Table 3) or centered at the mean within sub groups (i.e.

average QOR in high (5.6) or low (4.4) QOR subgroups in FEST, Paper I, Table 4;

Paper IV, Table 2). When centering the moderator QOR at the average level in the low QOR sub group, the model used is:

PFS = intercept + time + time x treatment + (QOR- 4.4) + time x treatment x (QOR-4.4) Thus centering is used to provide an interpretable or meaningful zero point.Time x treatment can be interpreted directly as treatment effects for the typical (average) low QOR patient, because the terms including QOR are zero for patients with QOR scale scores=4.4.

(31)

The moderator “gender” are coded as women=1 and men=0 or as women=0 and men=1.When women are coded as 0 (Paper III; Table 3 and Table 4) time x treatment can be interpreted as the treatment effect for women.

Ex.: Table 3 in Paper II: QOR is centered at the average level of QOR in the low QOR sub sample. Time is coded 1 for the transference group and 0 for the non-transference group and gender is coded 0 for women and 1 for men. The chosen QOR-level and women are set to be the zero-points. Time x treatment might therefore be interpreted as the treatment effect of transference interpretations for low QOR women.

2.5 Research on transference interpretations

More than 13.000 book chapters and articles have been published on the issue of transference (Search in PsychInfo/Ovid 30/6-09). Several questions have been debated.

What is the definition of transference interpretations? Is the transference a new experience or to which degree should the transference be viewed as an enactment of an earlier relationship (Piper et al., 1991; Section 2.2, present synopsis)? Which patients are best suited for these therapist interventions (Valbak et al., 2004)? What

recommendation can be proposed on frequency in time-limited psychotherapy?

Mainstream clinical thinking has maintained that transference interpretations are anxiety provoking and should therefore be used only for less disturbed and suitable patients in brief dynamic psychotherapy (Høglend, 1993; Piper et al., 1993; Høglend, 1996), although empirical research on this is contradictory (Høglend, 2004).

9 previous empirical studies attempting to shed light on the technical use of transference interpretations in brief dynamic psychotherapy have been published. None of these studies used an experimental design. The associations reported in passive observational studies are open to several causal interpretations, rendering the research base limited and inconclusive (Høglend, 2004). Up till now, the only controlled study using a quasi-experimental methodology, has been a previous study from Høglend and colleagues (1993). The quasi-experimental design was performed by selecting the patients to the different treatment based on suitability. Highly suitable patients (high QOR) received dynamic psychotherapy with a high number of transference

interpretations per session (on average 6). Less suitable patients (low QOR) received dynamic psychotherapy with few transference interpretations The authors reported that

(32)

the high QOR patients showed a significant negative effect during follow up compared to the low QOR patients. The study from Høglend and colleagues might have some limitations. Treatment integrity was not measured (Høglend, 1994). The sample size (transference group N=22, non transference group N= 21) was relatively low. Since a quasi-experimental design was used in the study, the two patient groups were unequal and selection maturation effects may have played a role.

2.6 Moderators in psychotherapy

Patient variables that influence treatment outcome are either predictors or moderators of outcome (Kraemer et al., 2002; Johansson and Høglend, 2007). Both predictors and moderators are pre-treatment variables that affect the strength or direction of a treatment response. Predictors do so regardless of treatment condition.

Moderators, however, differentially influence outcome depending on treatment condition.

It has been suggested that as much as 40 % of outcome variance in psychotherapy can be attributed to patient variables. This would mean that pre-treatment patient characteristics would account for the largest proportion of variance in psychotherapy outcome (Clarkin and Levy, 2004).

Differences between various psychological treatments have been difficult to detect and studies that have examined moderators of outcome have done so with limited success. That is, different treatments appear to be equally effective when looking at the average response. This could mean that there are no differences. Another possibility is that although treatments are equally effective on average, patients benefit differentially depending on pre-treatment characteristics. Sub groups of patients may respond very differently. Patient characteristics that are relevant to interpersonal processes might interact with different techniques used in treatment. To identify those pre-treatment variables might be helpful in understanding why different patients benefit more than others from different psychotherapy modalities.

2.6.1 Quality of Object Relations (QOR)

Melanie Klein and successors developed object relation theory (Klein, 1952).

However, inner personal patterns and past objects are not an easily measurable and

(33)

usable concept in psychotherapy research. The Quality of Object Relations scale (QOR) is one dimension supposed to predetermine suitability for dynamic psychotherapy (Høglend et al., 2000; Høglend, 2003; Azim et al., 1991). QOR is probably the best studied predictors in dynamic psychotherapy (Huprich and Greenberg, 2003). In the First Experimental Study of Transference-interpretations QOR was a pre-selected putative moderator in the study protocol.

QOR is measured on three 8-point scales: quality of interpersonal relationship in the patient’s life, history of adult sexual relationships, and history of nonsexual adult relationships (Appendix 8.2). QOR measures from primitive to mature the patient’s lifelong tendency to establish certain kinds of relationships with others (Azim et al., 1991).

Four empirical studies have investigated the association between level of QOR and use of transference interpretations. Two studies reported a negative effect of transference interpretations and outcome within the high QOR sub sample (Piper et al., 1991;

Høglend et al., 1993). Two other studies reported negative effects of transference interpretations within the low QOR sub sample (Connolly et al., 1999; Ogrodniczuk et al., 1999). Piper et al. (1991) and Høglend et al. (1993) reported high levels of transference interpretations per session (on average 5-6), Ogrodniczuk et al. (1999) reported a moderate level (on average 3.7 per session), and Connolly et al. (1999) reported a low level (on average 1 per session). The somehow contradictory findings may be explained by the frequency or level of interpretations, different patient samples, different therapists, and different measures. E.g. Connolly and associates evaluated the patient’s current level of quality of interpersonal relationships, while Piper and associates and Høglend and associates evaluated the life-long history of interpersonal relationships. It may also be the case, that since these studies are non-experimental, some of the correlations may be spurious (Høglend, 2004).

2.6.2 Patient gender in psychotherapy

In clinical theory, analysts recognize that patient gender contributes to the relationship between the therapist and the patient and to the patients’ thoughts about the therapist (Notman and Nadelson, 2004; Opdal, 2007; Kirschner et al., 1978;

Ogrodniczuk et al., 2004). Some authors (Friedman and Downey, 2008) claim that behavioral gender differences between women and men are core components of

(34)

knowledge about psychological development and should therefore be a core component of psychodynamic thinking and research as well. Most empirical studies and several research reviews, however, indicate that, on average, women and men respond similarly across different types of psychotherapy (Clarkin and Levy, 2004; Lam and Sue, 2001).

Those studies of individual psychotherapy have explored gender mainly as a general predictor. When the research group in FEST started investigation on patient gender and psychotherapy, only two studies had explored gender as a moderator. That is, they tested whether women and men respond differentially to different psychotherapies (Zlotnick et al., 1996; Ogrodniczuk et al., 2001).

Zlotnick et al. (1996) searched for predictor and moderator effects of gender in the National Institute of Mental Health Collaborative Study on Depression. The patients received short-term interpersonal therapy, cognitive-behavioral therapy, imipramine plus clinical management, or placebo plus clinical management. No significant predictor or moderator effects of patient gender were found at post-treatment. A study by Ogrodniczuk et al. (2001) showed a moderator effect of gender in two forms of individual psychotherapy at post-treatment. They reported that male patients had better outcomes in interpretive therapy than in supportive therapy, while female patients had better outcomes in supportive therapy compared to interpretive therapy. They did not find a moderator effect of gender at 1 year follow-up.

A recent study (Frank et al., 2008) compared the effect of combined

interpersonal and social rhythm therapy to clinical management. The authors reported that patients with Bipolar I Disorder made rapid gains in occupational functioning during and after combined interpersonal and social rhythm therapy compared to clinical management. The treatment effect was significantly larger in women than in men.

3 The present studies

3.1 The First Experimental Study of Transference-interpretations (FEST)

In the present dissertation data from The First Experimental Study of Transference-interpretations (FEST) were used. FEST is a dismantling randomized clinical trial, designed to explore specific long-term effects of transference

interpretations in dynamic psychotherapy. The study was initiated in March 1993, by a

(35)

research group led by professor Per Høglend. By December 2001 all hundred patients were included in the study. All follow-up evaluations were completed by December 2005. The patients were randomized to one of the two treatment groups. Both treatments employed use of general psychodynamic principles. One treatment (N=48) avoided an interpretive focus on the ongoing patient-therapist interaction (comparison group). The other treatment (N=52) used material from the patient-therapist interaction as the most important vehicle for clarifications, confrontations and interpretations (transference group). The design of the study is a RCT dismantling design, in which a single component (analysis of transference) is added or subtracted to an existent treatment package. Thus, the efficiency of a specific technique can be identified.

3.1.1 Main hypotheses

In FEST the primary hypothesis was that the transference group would have a more favorable course over the whole study period of 4 years compared to the

comparison group. The second hypothesis was that patients with mature object relations and/or absence of personality disorders would benefit more from therapy with

transference interpretations than from therapy without.

3.1.2 Post-hoc hypotheses

After reviewing the literature on gender and psychotherapy we found that no more than two empirical studies had explored the interaction effect of gender and treatment mode. Only one single study had found a moderator effect of patient gender and only during therapy, not at follow-up (Ogrodniczuk et al., 2001). The following was predicted: patient gender will not predict outcome across treatments, but women and men may respond differently during dynamic therapy with or without transference interpretations. Further exploratory analyses were made on gender differences during the whole study period controlled for the moderator effect of QOR.

(36)

3.2. Method

3.2.1 Patients

Patients from general practice, private specialist practices and psychiatric outpatient departments were referred to the study therapists and assessed for eligibility (N=122). Inclusion criteria were liberal. The patients sought psychotherapy for depressive disorders, anxiety disorders, personality disorders and interpersonal problems not caused by a mental disorder. Patients with psychosis, bipolar illness, organic mental disorder or substance abuse were excluded. One hundred patients were included and randomized to the two different dynamic psychotherapies (Figure 1, Paper I). The patients were unaware of the randomization and the technique studied. They were told that the aim of the study was long-term effects of dynamic psychotherapy.

3.2.2 Treatment

The patients were offered 45-minute weekly sessions for 1 year. All sessions were audio recorded. A treatment manual in Norwegian was published (Høglend, 1990).

All the patients in both treatment groups received psychodynamic psychotherapy. For the transference group, specific techniques were prescribed, including: the therapist addressed transactions in the patient–therapist relationship, encouraged exploration of thoughts and feelings about the therapy and the therapist, and interpreted direct manifestations of transference. Repetitive interpersonal patterns were linked to the transactions between the patient and the therapist. In the comparison group, these techniques were proscribed. In this group, the therapist consistently used material about interpersonal relationships outside of therapy as the basis for similar interventions (extra-transference interpretations), without any link to the interaction between the patient and the psychotherapist. Both treatments were mainly exploratory in nature.

Patients in both treatment groups were encouraged to explore sensitive topics, which often involved uncomfortable emotions, but the therapist abstained from providing guidance or giving advice, praise, or reassurance.

3.2.3 Therapists and evaluators

Patients were assigned to 1 of 7 therapists, depending on availability. The therapists, who also served as clinical evaluators of other patients, included 6